Question 1

What is this list?

Accepted Answer

The work pharma and biotech teams still do manually, written down. 742 recurring processes across ten departments, from discovery through commercial, each paired with the agent that takes it on. Every row says what a person does today, where the agent runs, and what stays with a human.

Question 2

What do "Drafts only", "Person approves" and "Unattended" mean?

Accepted Answer

They are the three levels an agent can run at, set per process. Drafts only: the agent prepares the work or gives the steps and a person does the rest. Person approves: the agent prepares and proposes, and nothing is written until a person signs off. Unattended: the agent runs on its own and queues exceptions for a person. GxP steps sit in the first two, and you can always set a process lower than shown.

Question 3

What are the eight agent types?

Accepted Answer

Document intake agents read documents and populate records. Monitoring agents watch for drift, deadlines and exceptions. Ask-your-data agents answer questions from your systems with the reference attached. System-of-action agents draft, route and write back across systems. Legacy-system wrappers work with systems and instruments that have no interface. Back-office agents reconcile and accrue. Chat and voice agents take input from a person, such as a rep's spoken debrief, and turn it into a record.

Question 4

What does "What stays human" mean on a card?

Accepted Answer

The step the agent never takes. For a deviation it is the QA decision, for a safety case it is the medical assessment, for a batch it is disposition. It is on every card so there is no ambiguity about where the agent stops and the accountable person starts. In practice each company sets this to its own rules: if your SOP requires a human at a step the card does not show, that step is enforced when the agent is set up.

Question 5

Is every use case here ready to deploy?

Accepted Answer

Every row is deployable. Some are ready today, others are built when you pick them, by Clientell AI's forward-deployed engineers. Either way the agent is fitted to your systems and your process before it runs, so what you get is never a generic version of the row.

Question 6

My process is not in the list. Can you still build it?

Accepted Answer

Yes. Name it and Clientell AI's forward-deployed engineers scope it, build it and fit it to your systems.

Question 7

Can we combine several use cases into one workflow?

Accepted Answer

Yes. Rows are written one process at a time so they are easy to find, but agents chain. Deviation triage, the investigation draft and the CAPA record are three rows and one flow in practice. Tell us the end-to-end outcome and the engineers assemble the chain.

Question 8

Which use case should we start with?

Accepted Answer

The one your team most wants off its plate that is not on the regulatory critical path. Reconciliations, chasing, intake, status reporting and drafting are typical first picks because the value shows quickly and the approval model is simple. Filter by "Unattended" or "Drafts only" to see the candidates.

Agents

Foundations

By Role

Use Cases

Library

Audits

Alternatives

Directories

Enterprise AI Harness

742 Agentic AI Use Cases for Life Sciences